World-model research
Explore how causal video generation accumulates state, responds to actions, and approximates physical dynamics over time.
Independent tool overview
Odyssey-2 at a glance
Odyssey-2 is an experimental world model that begins streaming an AI-imagined video world immediately and lets the user change what happens next with prompts.
Visit the official Odyssey-2 site ↗
- Product type
- Real-time interactive world model
- Initial release
- October 27, 2025
- Inputs
- Text or image to start; prompts to steer the stream
- Output
- Continuous AI-generated interactive video
- Published speed
- 20 frames per second; under 50 ms per frame
- Published duration
- Multi-minute streams and simulations
- Newer models
- Odyssey-2 Pro and Odyssey-2 Max
- Pricing
- No public self-serve price; API access by request
Overview
What Odyssey-2 is
Odyssey-2 is not a conventional text-to-video generator. Instead of rendering a fixed clip with a predetermined ending, it generates the next frame from what has happened so far and continues the simulation while the user interacts with it.
Odyssey says the model starts streaming immediately, produces a frame in under 50 milliseconds, runs at 20 frames per second, and can continue for minutes. Users can begin with text or an image and shape the evolving video with new prompts.
The original Odyssey-2 research experience remains available, but the model family has advanced. Odyssey-2 Pro launched with developer endpoints for simulations, interactive streams, and one-to-many viewable streams, followed by the larger Odyssey-2 Max. The current developer portal says Max access is rolling out to existing API users and asks new developers to request priority access.
The experience is a useful preview of real-time world simulation, interactive storytelling, training, games, and embodied-AI research. It is still early technology, with private pricing, limited production assurances, and legal terms that require careful review before commercial use.
Use cases
Who Odyssey-2 is best for
The strongest fit depends on the job you need the product to complete, not the size of its feature list.
Interactive media prototypes
Test experiences where a viewer can redirect a visual world while it is being generated.
Game and simulation experiments
Prototype open-ended environments without first building every scene and transition by hand.
Education and training concepts
Explore adaptive visual scenarios that change in response to a learner's questions or decisions.
Developers evaluating streaming world models
Use the API access process to assess interactive, viewable, or offline simulation endpoints for a product concept.
Capabilities
Core Odyssey-2 features
Instant streaming
Begins generating the visual stream immediately instead of waiting for a complete clip to render.
Real-time frame generation
Odyssey reports one new frame in under 50 milliseconds, corresponding to a 20-frame-per-second stream.
Multi-minute rollouts
Continues a simulation for minutes rather than stopping at a short fixed clip.
Open-ended interaction
Accepts new prompts while the video is running so the user can alter what the model generates next.
Text or image initialization
Starts a simulated world from a written prompt or an image according to Odyssey's current product overview.
Causal generation
Predicts each frame from prior frames and actions rather than relying on a known future ending.
General-purpose world knowledge
Aims to model many kinds of scenes, motion, lighting, contact, and behavior instead of one narrow environment.
Interactive Stream API
Odyssey-2 Pro introduced an endpoint for embedding a stream that can be changed programmatically in real time.
Viewable Stream API
Supports distributing one generated stream to multiple viewers for shared viewing experiences.
Simulation API
Generates an offline video from a prompt, specified actions at precise time steps, quality settings, and a target duration.
Developer SDKs
Odyssey announced JavaScript and Python SDKs with the Pro API, while current access is managed through its developer portal.
Evolving model family
Pro expanded the original experience into an API, and Max increases model scale and physical-simulation ambition.
Process
How the Odyssey-2 workflow works
Step 1
Choose research or developer use
Use the public Odyssey-2 experience for exploration; request API access when building an application or conducting structured evaluation.
Step 2
Check the current terms
Confirm allowed commercial use, data licensing, output handling, access limits, and any signed order before uploading sensitive material.
Step 3
Start with a clear world state
Describe the setting, subjects, visual style, camera viewpoint, and initial action, or provide an image you are authorized to use.
Step 4
Let the stream establish itself
Observe the first moments before adding instructions so later prompts respond to a stable visual context.
Step 5
Steer one change at a time
Use short, unambiguous prompts that change a subject, action, environment, or direction without overloading the model.
Step 6
Test long-horizon consistency
Look for drift in identity, geometry, physics, lighting, camera position, and cause-and-effect as the simulation continues.
Step 7
Choose the right API endpoint
Use interactive streams for live control, viewable streams for an audience, or simulations for action-timed offline output.
Step 8
Add application safeguards
Moderate prompts and outputs, constrain user actions, protect API credentials, disclose generated content, and provide failure recovery.
Step 9
Measure before scaling
Benchmark response time, stream stability, quality, safety, cost, and the percentage of sessions that produce a useful experience.
Cost
Odyssey-2 pricing and free plan
Odyssey does not publish a standard consumer subscription or per-second API rate. The current developer portal asks new users to request priority access, and API fees are set through the applicable order. Treat any trial or promotional access as temporary and obtain a written quote before budgeting.
Odyssey-2 research experience
Price not published
The public-facing experience for trying the original interactive model.
- Access conditions can change as the research service evolves
- The consumer terms restrict commercial use
- Do not assume the experience includes production rights or service guarantees
Developer API priority access
Request access
The developer portal is currently onboarding new users through a priority-access request.
- Odyssey-2 Max is rolling out to existing API users
- Usage limits and rate limits are defined in documentation or communicated by Odyssey
- API access can be suspended or throttled under the agreement
Commercial API order
Custom quote
Fees and permitted use are defined by an API order and governing agreements.
- Unless an order says otherwise, fees are billed monthly in advance
- The standard order language uses month-to-month renewal
- Confirm support, service levels, data terms, overages, and application rights in writing
Pricing checked . Check current pricing at the source ↗
Assessment
Odyssey-2 strengths and limitations
Where it stands out
- Generates an evolving stream instead of a fixed clip with a predetermined ending
- Published 20-frame-per-second speed supports immediate interaction
- Multi-minute output creates a different design space from short video generators
- Natural-language steering makes the research experience approachable
- Text and image initialization support different starting points
- Interactive, viewable, and simulation endpoints cover live and offline developer workflows
- The Pro and Max releases show rapid progress beyond the original model
- Well suited to experimenting with emergent media, adaptive worlds, and agent environments
What to consider
- Odyssey describes the model family as nascent, so it should be treated as experimental rather than a dependable world simulator
- Errors can compound across long causal rollouts and cause identity, geometry, physics, or camera drift
- Real-time speed does not guarantee the visual fidelity or controllability of slower fixed-clip generators
- The public experience is not a precise video editor with a timeline, shot locking, or deterministic revisions
- The current developer portal uses priority access rather than unrestricted self-serve onboarding
- No public standard price makes cost comparison and forecasting difficult
- The standard API order provides no support, service levels, uptime, or maintenance commitments unless separately agreed
- The consumer app terms limit use to personal, non-commercial purposes and separately prohibit commercial use of the service
- The prototype API license is limited to internal business application development unless Odyssey grants additional written rights
- Odyssey's terms grant it a broad, perpetual license to use prompts and outputs, including for model training and improvement
- Content posted publicly can be viewable, modifiable, and remixable by other users
- Outputs may be inaccurate, undesirable, non-unique, or similar to results generated for others
- Users remain responsible for intellectual property, privacy, publicity, accuracy, and legal compliance
- The service's content rules restrict adult, violent, political-campaign, deceptive, and other sensitive uses
Compare
Odyssey-2 alternatives
The right alternative depends on the specific output, workflow, controls and budget your project requires.
Content Creator
World Labs Marble
A stronger fit for creating persistent, navigable 3D worlds from text, images, or video rather than a continuously generated 2D video stream.
Explore World Labs Marble →Content Creator
Veo 3.1
A conventional high-quality video model for directed, fixed clips with production-oriented controls and native audio.
Explore Veo 3.1 →Content Creator
Runway Gen-4.5
A creator-focused video model and editing ecosystem when cinematic clip quality and repeatable production tools matter more than live world interaction.
Explore Runway Gen-4.5 →Questions
Odyssey-2 FAQs
What is Odyssey-2?
Odyssey-2 is a causal world model that generates a continuous video stream frame by frame and lets the user change what happens next with prompts.
How is Odyssey-2 different from a normal AI video generator?
Most video models render a fixed clip. Odyssey-2 starts streaming immediately, predicts the next frame from prior context, and can respond to new instructions while the simulation continues.
How fast is Odyssey-2?
Odyssey reports that it generates a frame in under 50 milliseconds and streams at 20 frames per second.
How long can Odyssey-2 run?
Odyssey describes multi-minute interactive streams and simulations rather than the short fixed clips common in video generators.
Can I start from an image?
Yes. Odyssey's current product overview says Odyssey-2 can begin from an image or a text prompt.
What are Odyssey-2 Pro and Odyssey-2 Max?
Pro extended the original Odyssey-2 into a developer API with interactive and offline endpoints. Max is the newer, larger model now rolling out to existing API users.
Is Odyssey-2 available through an API?
Yes, but current access is managed. Existing API users are receiving Max access, while new developers are directed to request priority access.
What API modes are available?
Odyssey announced interactive streams for live control, viewable streams for one-to-many distribution, and simulations for offline videos with actions scheduled at specific times.
How much does Odyssey-2 cost?
Odyssey does not publish a standard self-serve price. API fees are defined in an applicable order, so developers need access approval and a current quote.
Can I use the public Odyssey-2 experience commercially?
The current consumer terms restrict the app and service to personal, non-commercial use. Commercial or product work should use an appropriately authorized API agreement and written order.
Who owns Odyssey-2 output?
The terms say users retain their rights in output subject to law, but they also grant Odyssey a broad perpetual license to use user content and output, including for service operation and model improvement.
Is Odyssey-2 suitable for production applications?
It is promising for prototypes and research, but production teams should first secure the correct license, pricing, support terms, safety controls, reliability evidence, and a fallback for unstable streams.
Bottom line
Our Odyssey-2 verdict
Odyssey-2 is one of the clearest demonstrations of how AI video can become an interactive medium rather than a rendered asset. Immediate, multi-minute generation is genuinely distinctive, and the Pro and Max roadmap makes it relevant to developers as well as researchers. It remains an early platform: use it to test new interaction ideas, but do not treat the public experience as a production-ready commercial video tool.
Visit Odyssey-2 website ↗Similar Content Creator Tools
View all Content Creator AI tools →
Sesame
Sesame - Conversational voice agent capable of holding natural conversations
Hailuo 2.3
Hailuo 2.3 - MiniMax's new AI video model with upgraded movement, realism, and expression
Spiral
Spiral - An AI writing partner with taste
Sonic 3
Sonic 3 - Cartesia's realistic text-to-speech model for voice agents
AI training for the future of work.
Get access to all our AI courses, hundreds of real-world AI use cases, live expert-led workshops, an exclusive network of AI early adopters, and more.
AI Courses
Get unlimited access to all of our current & upcoming industry-specific AI courses for the duration of your subscription.
Daily Guides
To keep up with the rapid pace of AI, our team publishes AI implementation guides daily. Our library contains 300+ practical use cases to automate real-world work.
Workshops
Join weekly, live, interactive sessions with industry leaders who are at the forefront of AI for hands-on implementation guidance and exclusive insights.
Community
Network with an exclusive community of AI-first professionals who are working smarter with AI. Learn how early adopters are using AI in their work and businesses.